There’s a Thunderbolt 4 bottleneck of 32Gb/s (which uses 4x PCIe 3 lanes). The 4x PCIe 4 lanes would double the speed (of TB4 / 4x PCIe 3) to 64Gb/s.
Pretty sure this is what TB5 is meant to do, but if this can be done manually without needing to implement a non-existent specification, that’d be cool. I’ve no clue how.
4x PCIe 4 lanes should be enough to run most GPUs at some minor bottleneck - much better than my current TB4->TB3, 4x PCIe 3.
Before putting my laptop in a backpack, I will usually enter it into sleep mode (Win 10) either via hitting the power button or start menu. Sleep is preferable vs fully powering off to persist application states.
Entering Sleep seems to have a ~10s delay before entering Sleep (power led sleep animation starting) during which any mouse or keyboard input will wake the machine aborting the Sleep process. Plugging or unplugging USB-C (power) seems to also wake the machine.
Most of the time, this flow works out okay if I wait to see the LED sleep status activate. Occasionally I’ll find the machine awake in my bag when unintended. Some discussion in a forum thread here.
I wonder if a hardware switch might be a better UX- set the switch to Sleep and it’s out 100% of the time (suppresses any other wake mechanisms) until the switch is toggled. Not sure if this is technically possible over USB, though if it is could perhaps make a expansion card including both a sleep switch and a USB port?
How come you do not have it set to sleep when the lid is closed?
I have an interesting idea for an expansion card (which I’d like to note I have yet to see it on the forum but I’m not claiming I’m the first with this idea). After the revelation of extended expansion cards being a thing that can happen(i previously never considered it), I had an idea for one that could house an M.2 2230 SSD. I have not done any cad work or modeling with this, it just came to mind and I’d thought I should share it. It seems to be like there could be promise in doing some sort of internal USBC to M.2, then being able to slot your own drive in through a removable lid on the card.
I will say that it does seem a little unnecessary since there are already drive expansion cards, I just think it would be interesting for modularity and compatibility with other systems. Anyone have any thoughts on this?
Would want to take this on but thermals and silicon shortages will be your issue as I know the silicon required are most of the time either very expensive or out of stock. /
When you say “extended expansion cards”, do you mean double-height, or extending past the side of the laptop?
I wonder if a double-height card could house a 2280 or even 22110 card… it would have to extend further “in” than the normal “back” of the slot, but if it’s under the body, that might be feasible? (For that matter, you could steal some additional room to the sides also, as long as you don’t block the release tab…)
@matthew3 Initially I only considered it extending out of the side of the laptop, but not increasing the height. Increasing the height sounds mildly interesting, or at least lengthening the side with the port on it with more room for larger ports(ethernet??). I imagine that a card to fit something like a 2280 or 22110 would have to extend under the laptop inwards. I like that idea a lot actually, I wonder if it is something they have considered.
I’ve enquired to a few manufacturers of the silicon required to create an expansion card like this, if I am able to get datasheets and purchasing options without an NDA I will continue with this idea.
Edit: misread the OP as 2280…
is m.2 hot swappable?
by spec I think so
–padding–
Nvme ssd dock
If you want PCIe, the most obvious solution is to leverage thunderbolt. Intel is the only manufacturer of thunderbolt controllers. And their datasheets are… nothing but product briefs. I suspect you need to sign an NDA with them to get the full datasheet and documentations.
USB4 can be a way out as it supports native PCIe tunneling. But it is an optional feature by the spec. I am not sure if Framework Laptop supports it. Plus the only USB4 controller (ASM2464PD from ASMedia) that has announced support is still a demo. Rumor has it that it will be shipped in Q4 this year.
I have no idea if it’s possible with the framework but external gpu’s have been done would they work? i don’t know if you would design your own system or see about supporting an existing system “would probably need to check for licensing issues”
Like dell’s.
As framework supports thunderbolt (unofficially as they didn’t get the licenses yet), eGPU’s are indeed already supported. I think there are threads about it where people tried it out and shared their experience.
Thunderbolt 3, m.2, and mPCIe eGPUs will all work with the framework, best bandwidth/convenience trade off being Thunderbolt 3. In my case, purchasing an x16-to-m.2 board with an m.2-to-thunderbolt converter, a power supply, and an old midrange graphics card roughly doubled the cost of my setup, but graphics card prices have come down significantly since then.
Windows works flawlessly; Linux has certain teething problems.
My setup is Framework 11th gen EndeavourOS (arch derivative) linux + Sonnet e550 + nVidia 1050Ti . This works well on X11 with some manual configuration, but in Wayland more problems by far.
So yes, the hardware is perfectly capable of supporting eGPUs. This has been well documented in other threads in this forum. Search is your friend.
Nothing popped up searching for external gpu.
But I didn’t know they get called egpu’s
that’s because they’re a standard desktop gpu in an enclosure that provides an interface and power. You’d generally get the enclosure and the gpu separately.
You can find a lot more info than is available here on the forum egpu.io. Anything on there that’s Thunderbolt 3 or 4, m.2, or mPCIe will work with the Framework (the preference being Thunderbolt for many reasons you can find there as well).